OpenAI has made GPT-5.3 the default model in ChatGPT for logged-in users, with GPT-5.3 Instant rolling out across accounts over the next few days, according to the company’s Help Center documentation.
OpenAI said the update is an upgrade aimed at day-to-day usefulness, not a headline benchmark jump, saying the update targets issues that “don’t always show up in benchmarks” but still determine whether ChatGPT feels “helpful or frustrating,” in a product post.
GPT-5.3 Instant is, though, “default off” in ChatGPT Enterprise and Edu workspaces unless an admin enables it, while consumer users see it as the default in everyday use.
What OpenAI says changed
OpenAI said GPT-5.3 Instant reduces “unnecessary refusals” and tones down “overly defensive or moralizing preambles,” after feedback that GPT-5.2 Instant sometimes declined safe requests or led with lengthy caveats.
The company also said it changed how the model uses web results. GPT-5.3 Instant is designed to better balance online information with its own reasoning, including using existing context to interpret recent news instead of returning loosely connected summaries.
On factual accuracy, OpenAI said it measured hallucinations in two internal evaluations: a high-stakes test covering domains such as medicine, law and finance and a separate user-feedback evaluation based on de-identified ChatGPT conversations flagged as factual errors.
In those tests, OpenAI said hallucinations fell by 26.8% when using the web and 19.7% when relying only on its internal knowledge, compared to prior models. OpenAI also reported hallucinations fell by 22.5% with web use and 9.6% without web access on the user-feedback evaluation.
Why the default model shift is a strategic signal
OpenAI described GPT-5.3 Instant as an update to ChatGPT’s “most-used model,” which makes this update OpenAI’s proving ground for refusal tuning, web answers and accuracy in day-to-day use.
The Help Center says ChatGPT’s Auto setting can switch between GPT-5.3 Instant and GPT-5.2 Thinking depending on the task.
For enterprise teams, OpenAI’s change log is more about production friction points than new benchmark ceilings: refusals, web answer quality and reduced hallucinations.
The upgrade is available immediately to all ChatGPT users and to developers via the API under the alias gpt-5.3-chat-latest. In its API documentation, OpenAI describes gpt-5.3-chat-latest as pointing to the GPT-5.3 Instant snapshot used in ChatGPT, with a listed 128,000 context window and pricing listed per 1M tokens for text input, cached input and output.
GPT-5.2 Instant will remain available for paid ChatGPT users in the Legacy Models section for three months, after which it will be retired on June 3, 2026.
The AI firm also teased GPT-5.4 on X, posting “5.4 sooner than you Think” from its official account, without providing dates or product details.